Subpixel edge location using improved LSR

Abstract

Subpixel methods for edge location are of significant importance in computer vision and other applications. Such methods often use local surface models or polynomial approximation for a local neighborhood to obtain the subpixel edge location. Whereas their performance are good in smooth area, they are typically poor to deal with singularities presented in discrete data and bring discontinuous edge location. In this paper, we propose an improved level set reconstruction(ILSR) technique to overcome these limitations by locating curves while: 1) obtaining continuous and smooth curve, and 2) keeping edge, especially corner location precision. This is accomplished by ILSR scheme in which curves are smoothed along isophotes with anchor and topology constraints at corners which are guided by corner judgement This method results in subpixel location with accurate edge position.